[BioC] limma help - choosing an approach
Naomi Altman
naomi at stat.psu.edu
Fri Sep 15 04:05:55 CEST 2006
I don't think that the single channel analysis is experimental. It
follows the rules of mixed model analysis which has been around for a
very long time.
Any set of contrasts is statistically valid. The question is which
ones may be biologically meaningful. And that depends in part on the
biological process.
E.g. often contrasts among time points with the same concentration
and concentrations with the same time point may be of most
interest. But if the chemical slows the cell metabolism by a factor
of 4, you might want to do contrasts of concentration 1 time 1 with
concentration 2 time 4 etc.
--Naomi
At 06:19 AM 9/14/2006, you wrote:
>Hello Naomi (and anyone else)
>
>Thanks again for your help. It has been extremely helpful. Almost there
>I think, though I am still just doing it on autopilot and not quite sure
>how meaningful this is. I am slightly worried by the limma documentation
>which describes the single channel analysis as "experimental" so
>combined with my lack of understanding makes me feel quite uncertain.
>Anyway I can run it now and pull out lists of genes - though what they
>mean I am not sure yet.
>
>So, a bit of sanity checking if that is OK.
>
> > a) I needed to build my targets from a list of treatments for red and
>gree.
>
>I guess these are equivalent to Cy5 and Cy3 with Genepix. The function
>"targetsC<-targetsA2C(targets)" converts it to single channel format.
>(See targets files below, showing before and after).
>
>
> >b) The design matrix will have more than 4 columns, as you have 3
> >levels of concentration. So there are 2 columns for concentration
> >and 2 for time:concentration.
>
>I have altered the targets file to reflect this better.
>
>So now my code looks like:
>
>Time<-targetsC$Time
>Conc<-targetsC$Target
>Time<-factor(Time, levels=c("t1", "t4"))
>Conc<-factor(Conc, levels=c("c0", "c20", "c100"))
>
>
>Giving:
>
> > Time
> [1] t1 t1 t4 t4 t4 t4 t1 t1 t1 t1 t4 t4 t1 t1 t1 t1 t4 t4 t4 t4 t4 t4
>t1 t1 t4 t4 t1 t1 t1 t1
>Levels: t1 t4
> > Conc
> [1] c100 c0 c20 c0 c100 c0 c20 c0 c20 c0 c20 c0 c20 c0
>c100 c0 c20 c0 c20 c0 c100 c0 c100 c0 c100 c0
>[27] c100 c0 c20 c0
>Levels: c0 c20 c100
> >
>
>Set up the design file:
>
>design<-model.matrix(~Time + Conc + Time:Conc)
>
>Giving (first three lines):
>
> (Intercept) Timet4 Concc20 Concc100 Timet4:Concc20 Timet4:Concc100
>1 1 0 0 1 0 0
>2 1 0 0 0 0 0
>
>
>colnames(design)<-c("Intercept", "t4", "c20", "c100", "t4c20", "t4c100")
>
>
>corfit<-intraspotCorrelation(MA.nba, design)
>fit<-lmscFit(MA.nba, design, correlation=corfit$consensus)
>
>
># contrast matrix
>contrast.matrix<-makeContrasts(t4, c20, c100, t4c20, t4c100,
>levels=design)
>
>(What is possible here? What sort of contrasts are valid/meaningful?
>Presumably I could do c20 + c100 to compare against c0?).
>
>
>
># contrast timepoints and controls
>fit2<- contrasts.fit(fit, contrast.matrix)
>
>
># eBayes
>eb<- eBayes(fit2)
>
>
>ngenes<-20
>topa1<-topTable(eb, coef=1, number=ngenes, adjust="none", sort.by="M")
>
>.....
>
>
>
>
>Regards
>
>
>John Seers
>
>
>
>Targets data before transforming using targetsA2C
>
>
>SlideNumber FileName Cy3 Cy5 Time
>598 598new.gpr c100 c0 t1
>599 599new.gpr c20 c0 t4
>600 600new.gpr c100 c0 t4
>617 617new.gpr c20 c0 t1
>621 621new.gpr c20 c0 t1
>637 637new.gpr c20 c0 t4
>638 638new.gpr c20 c0 t1
>639 639new.gpr c100 c0 t1
>748 748new.gpr c20 c0 t4
>751 751new.gpr c20 c0 t4
>833 833new.gpr c100 c0 t4
>835 835new.gpr c100 c0 t1
>836 836new.gpr c100 c0 t4
>957 957new.gpr c100 c0 t1
>958 958new.gpr c20 c0 t1
>
>Targets data after transforming using targetsA2C
>
>
> > targetsC
> channel.col SlideNumber FileName Time Target
>598new.1 1 598 598new.gpr t1 c100
>598new.2 2 598 598new.gpr t1 c0
>599new.1 1 599 599new.gpr t4 c20
>599new.2 2 599 599new.gpr t4 c0
>600new.1 1 600 600new.gpr t4 c100
>600new.2 2 600 600new.gpr t4 c0
>617new.1 1 617 617new.gpr t1 c20
>617new.2 2 617 617new.gpr t1 c0
>621new.1 1 621 621new.gpr t1 c20
>621new.2 2 621 621new.gpr t1 c0
>637new.1 1 637 637new.gpr t4 c20
>637new.2 2 637 637new.gpr t4 c0
>638new.1 1 638 638new.gpr t1 c20
>638new.2 2 638 638new.gpr t1 c0
>639new.1 1 639 639new.gpr t1 c100
>639new.2 2 639 639new.gpr t1 c0
>748new.1 1 748 748new.gpr t4 c20
>748new.2 2 748 748new.gpr t4 c0
>751new.1 1 751 751new.gpr t4 c20
>751new.2 2 751 751new.gpr t4 c0
>833new.1 1 833 833new.gpr t4 c100
>833new.2 2 833 833new.gpr t4 c0
>835new.1 1 835 835new.gpr t1 c100
>835new.2 2 835 835new.gpr t1 c0
>836new.1 1 836 836new.gpr t4 c100
>836new.2 2 836 836new.gpr t4 c0
>957new.1 1 957 957new.gpr t1 c100
>957new.2 2 957 957new.gpr t1 c0
>958new.1 1 958 958new.gpr t1 c20
>958new.2 2 958 958new.gpr t1 c0
> >
>
>
>
>
>---
>
>John Seers
>Institute of Food Research
>Norwich Research Park
>Colney
>Norwich
>NR4 7UA
>
>
>tel +44 (0)1603 251497
>fax +44 (0)1603 507723
>e-mail john.seers at bbsrc.ac.uk
>e-disclaimer at http://www.ifr.ac.uk/edisclaimer/
>
>Web sites:
>
>www.ifr.ac.uk
>www.foodandhealthnetwork.com
Naomi S. Altman 814-865-3791 (voice)
Associate Professor
Dept. of Statistics 814-863-7114 (fax)
Penn State University 814-865-1348 (Statistics)
University Park, PA 16802-2111
More information about the Bioconductor
mailing list